Nvidia has agreed to guarantee just under $120 billion for OpenAI's massive data center build-out — nearly half the $250 billion originally planned — after investors pushed back on the chipmaker's risk exposure, the Wall Street Journal reported on Friday.

The scaled-back guarantee, confirmed in subsequent Reuters coverage, covers the first construction phase of the project, which would deliver about five gigawatts of computing capacity. It is one of the largest financial commitments in the history of the technology industry, even in reduced form. For more context on this story, see our ongoing artificial intelligence updates.

What changed in the deal

Under the original framework, Nvidia would have backed $250 billion of the data center development. Under the revised terms reported by the Journal, the company would initially guarantee just under $120 billion — a cut of more than 50 percent.

According to the report, the reduction was driven by investors, who raised concerns about the scale of Nvidia's risk exposure to a single customer. Nvidia's shares have been sensitive to perceived circularity in AI financing arrangements, in which chip vendors effectively fund the purchases of their own products.

The guarantee is not Nvidia's only financial entanglement with OpenAI. The company is also in talks over separate financing for OpenAI's chip purchases worth up to $350 billion, the Journal reported. The Next Wall Street noted that the backstop reduction surfaced in the same week that new 13F filings revealed the stakes Nvidia holds across its AI investments.

The 10-gigawatt question

The larger picture is even more ambitious than the first phase. OpenAI is separately negotiating a lease for the full 10-gigawatt project being developed by SB Energy, a subsidiary of SoftBank, according to the Journal's reporting.

If completed at that scale, the site would rank among the largest data center campuses ever constructed, built specifically to train and run frontier AI models. The five gigawatts covered by Nvidia's reduced guarantee represent only the opening tranche of that vision.

The agreement also deepens the web of financial relationships binding the AI industry's key players. OpenAI has already secured landmark commitments this year — including a $50 billion investment from Amazon that closed with the e-commerce giant taking an equity stake, as AI Buzz Wire previously reported — while SoftBank, Oracle, Microsoft and Nvidia have all signed compute or financing arrangements tied to the ChatGPT maker's expansion. Reuters characterized the project at the center of Friday's report as an Ohio data center development.

The circular financing problem

What made the original guarantee contentious was its structure. When a chip supplier guarantees financing for its own customer's data centers, the arrangement effectively converts Nvidia's product sales into Nvidia-backed credit — revenue today secured by risk tomorrow. For as long as AI demand keeps compounding, the loop is self-reinforcing. If demand ever stalls, the exposures land on the same balance sheet.

That is precisely the scenario Nvidia's shareholders appear to have priced in. The Journal's account of investor pushback suggests the company's own base was unwilling to carry $250 billion of concentrated customer risk, even for a customer as central to its growth as OpenAI. Halving the initial guarantee to just under $120 billion preserves the commercial relationship while capping the worst-case exposure.

Fuel for the bubble debate

The timing of the cut is what gives it significance beyond the two companies involved. Nvidia halving its risk under investor pressure will be read by critics as evidence that even the biggest beneficiaries of the AI boom are growing more cautious about its durability.

The report landed within hours of disclosures that pull the other way. Reuters reported that Anthropic's revenue more than doubled to over $11.5 billion in the second quarter, with the company projecting $190 billion to $200 billion in revenue for 2028 ahead of a potential IPO. As The Decoder observed, the investor pressure on Nvidia and Anthropic's surging numbers represent the two competing readings of the AI economy: one in which financing risk is catching up with the sector, and one in which demand keeps outrunning every forecast.

Why it matters

For months, analysts have warned that AI infrastructure spending has outpaced the revenue it generates. Nvidia's willingness to back $250 billion of OpenAI's build-out was one of the strongest corporate signals that the industry believed in its own demand projections. A guarantee cut roughly in half — under explicit shareholder pressure — softens that signal.

At the same time, a $120 billion commitment is hardly a retreat. The deal, if finalized, would still anchor one of the largest data center projects on the planet and lock in years of demand for Nvidia's AI accelerators.

What comes next

  • Final terms. The Journal reports the two companies are closing in on the data center deal, meaning final numbers could be announced within weeks.
  • The chip financing talks. The separate negotiation over up to $350 billion in chip purchase financing would dwarf the data center guarantee if completed on those terms.
  • SB Energy's lease. OpenAI's negotiation for the full 10-gigawatt campus will determine whether the project expands beyond the first phase.
  • Market reaction. Investors have punished perceived overexposure to AI financing risk this quarter, and Nvidia's disclosure schedule will be watched for further detail on the reduced guarantee.

For now, the AI build-out continues — just with its biggest sponsor a little less all-in than before.

Sources: Wall Street Journal, Reuters, The Decoder, The Next Wall Street.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →